Mechanisms of hematin crystallization and ... - Semantic Scholar

2 downloads 0 Views 2MB Size Report
Apr 21, 2015 - spontaneous formation of beta-haematin (malaria pigment). .... synthetic haemozoin (beta-haematin) nucleated at the surface of lipid particles.
Mechanisms of hematin crystallization and inhibition by the antimalarial drug chloroquine Katy N. Olafsona, Megan A. Ketchuma, Jeffrey D. Rimera,1, and Peter G. Vekilova,b,1 a Department of Chemical and Biomolecular Engineering, University of Houston, Houston, TX 77204; and bDepartment of Chemistry, University of Houston, Houston, TX 77204

Edited by Lara A. Estroff, Cornell University, Ithaca, NY, and accepted by the Editorial Board February 17, 2015 (received for review January 16, 2015)

Hematin crystallization is the primary mechanism of heme detoxification in malaria parasites and the target of the quinoline class of antimalarials. Despite numerous studies of malaria pathophysiology, fundamental questions regarding hematin growth and inhibition remain. Among them are the identity of the crystallization medium in vivo, aqueous or organic; the mechanism of crystallization, classical or nonclassical; and whether quinoline antimalarials inhibit crystallization by sequestering hematin in the solution, or by blocking surface sites crucial for growth. Here we use time-resolved in situ atomic force microscopy (AFM) and show that the lipid subphase in the parasite may be a preferred growth medium. We provide, to our knowledge, the first evidence of the molecular mechanisms of hematin crystallization and inhibition by chloroquine, a common quinoline antimalarial drug. AFM observations demonstrate that crystallization strictly follows a classical mechanism wherein new crystal layers are generated by 2D nucleation and grow by the attachment of solute molecules. We identify four classes of surface sites available for binding of potential drugs and propose respective mechanisms of drug action. Further studies reveal that chloroquine inhibits hematin crystallization by binding to molecularly flat {100} surfaces. A 2-μM concentration of chloroquine fully arrests layer generation and step advancement, which is ∼104× less than hematin’s physiological concentration. Our results suggest that adsorption at specific growth sites may be a general mode of hemozoin growth inhibition for the quinoline antimalarials. Because the atomic structures of the identified sites are known, this insight could advance the future design and/or optimization of new antimalarials.

|

malaria parasites heme detoxification chloroquine crystal growth inhibition

|

| crystallization mechanisms |

of hematin crystallization and its inhibition by antimalarials may prove to be influential for drug development (14). Despite many years of effort (7, 9, 12, 15–18), fundamental questions regarding the mechanism of hematin crystallization and its inhibition remain elusive. Among them are (i) What is the nature of the environment within the parasite where hemozoin crystals recruit hematin and grow? The two likely candidates are the aqueous phase in the parasite digestive vacuole (DV) (18, 19) and the lipid subphase that has been reported to reside either in the DV bulk (9, 16) or along the DV membrane (18–20). (ii) What is the mechanism of hematin crystallization?—classical, i.e., addition of molecules to growth sites (21–23), or nonclassical, i.e., association of precursors (24–26)? (iii) What is the mechanism of action of the inhibitor species? It is possible that the inhibitors either reduce the concentration and activity of hematin in the growth medium through complexation (27, 28), or interfere with crystallization by binding to the crystal surface(s) and restricting solute addition (23). The answers to these questions offer an improved understanding of malaria parasite physiology and may potentially lead to the rational design of hematin crystallization inhibitors that could serve as effective antimalarial drugs. As a model of hematin (Fig. 1A) crystallization we use the growth of β-hematin, the synthetic form of hemozoin. β-Hematin has a crystal structure (P1 symmetry) and habit identical to its natural analog (7), with predominant growth along its ~ c direction, (Fig. 1B). Both natural and synthetic hematin crystals assemble as high-aspectratio parallelogram-shaped platelets, with basal {100} faces and sides defined by {010}, (011), and (001) surfaces (17, 29). Significance

W

hereas significant public health initiatives have eradicated malaria from North America, Europe, and other developed regions of the world (1), the disease remains endemic in the equatorial regions of Africa, South America, Southeast Asia, and Oceania (2). Approximately 40% of the global population is at risk for malaria infection, predominantly from the protozoan parasite Plasmodium falciparum (2). Very disturbingly, a resurgence of the disease throughout the world has occurred since the 1960s due to the emergence and spread of Plasmodium parasites resistant to chloroquine combination treatments (2, 3). Delayed parasite clearance has been recorded for even the most recent artemisinin-based therapies (4). The weak responses to the common antimalarial drugs underscore the urgent need for research into the critical processes of malaria parasite physiology. Malaria parasites residing in the erythrocytes catabolize hemoglobin and release Fe(II) heme (5). The released heme rapidly oxidizes to toxic Fe(III) hematin, which is sequestered as crystalline hemozoin (6, 7). The traditional Western treatment for malaria, quinine, and its synthetic homologs (chloroquine, mefloquine, and others) (8–11) putatively works by blocking hematin crystallization (12). Available evidence suggests that artemisinin, another antimalarial drug, binds to heme (2, 13). The sequestration of heme into hemozoin is a suitable target for new antimalarials. Hence, an understanding of the mechanisms

4946–4951 | PNAS | April 21, 2015 | vol. 112 | no. 16

Approximately 40% of the global population is at risk for malaria infection and 300–660 million clinical episodes of Plasmodium falciparum malaria occur annually. During the malaria parasite lifecycle in human erythrocytes, heme released during hemoglobin catabolism is detoxified by sequestration into crystals. Many of the common antimalarials are believed to suppress the parasite by inhibiting hematin crystallization. We present, to our knowledge, the first evidence of the molecular mechanisms of hematin crystallization and antimalarial drug action as crystal growth inhibitors. These findings enable the identification and optimization of functional moieties that bind to crystal surface sites, thus providing unique guidelines for the discovery of novel antimalarials to combat increased parasite resistance to current drugs. Author contributions: K.N.O., J.D.R., and P.G.V. designed research; K.N.O. and M.A.K. performed research; K.N.O., J.D.R., and P.G.V. analyzed data; and J.D.R. and P.G.V. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. L.A.E. is a guest editor invited by the Editorial Board. 1

To whom correspondence may be addressed. Email: [email protected] or vekilov@ uh.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1501023112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1501023112

Results and Discussion

Olafson et al.

Fig. 1. β-Hematin crystals. (A) Structure of hematin. (B) AFM image of a β-hematin crystal on a glass substrate reveals a morphology similar to hemozoin crystals isolated from P. falciparum. (Scale bar, 2 μm.) (C) A 3D AFM height image of a (100) face reveals the presence of unfinished layers. The step height h = 1.17 ± 0.07 nm was determined by averaging measurements from multiple images. (D) Molecular model of β-hematin using the software package Diamond illustrates an unfinished layer (C atoms in white) on a (100) face (C atoms in blue).

of the quinoline antimalarials in organic liquids, which would seemingly reduce their efficacy toward crystals growing in a lipid phase. To this end, we determined the solubility of chloroquine (CQ), a common antimalarial drug, in CBSO and citric buffer at pH 4.8 (Fig. S6B). The solubility of CQ in CBSO is 0.19 mM, which is ∼104× less than its solubility in the aqueous solvent, but it is still twofold higher than the solubility of hematin in CBSO (∼0.1 mM). Collectively, these measurements suggest that the CQ solubility in CBSO is sufficient for growth inhibition by surface binding, a mechanism discussed in greater detail below. These arguments advocate that the lipid structures in the DV may be a preferred environment for hematin crystallization. The difficulty in crystal growth from a physiologically relevant aqueous environment may be attributed to hematin’s low solubility (2 nM) (34) or its propensity to form oligomers (36, 37) that could potentially adhere to the crystal surface and slow or block its growth. Even if one accepts that the crystals are not suspended in lipid nanospheres located in the DV bulk, but are attached to the DV membrane so that only one of the basal faces is exposed to lipids lining the DV membrane (18–20), this contact may be sufficient to ensure growth of the physiological hematin crystals. The Crystallization Mechanism in CBSO. Guided by the conclusion of preferential β-hematin growth in organic solvents, we used CBSO supersaturated with hematin as a growth medium and focused on the {100} faces of β-hematin. We used large β-hematin crystals prepared in the biomimetic CBSO solutions discussed above and performed, to our knowledge, the first time-resolved in situ AFM study of hematin crystal growth. In situ AFM has proven to be a valuable technique for elucidating structural and dynamic characteristics of classical and nonclassical crystallization mechanisms (38–41). AFM topographical images (Fig. 1C) reveal the presence of unfinished layers on a (100) face with heights h = 1.17 ± 0.07 nm, close to the unit cell dimension in the [100] direction (a = 1.22 nm, Fig. 1D) (7). PNAS | April 21, 2015 | vol. 112 | no. 16 | 4947

CHEMISTRY

Aqueous or Organic Medium for Hematin Crystallization. The parasite DV presents a complex environment. The vacuole comprises membrane interfaces (20), an acidic aqueous solution with pH 4.8−5.5 (30), and lipids, mostly mono- and diglycerides, resulting from the degradation of the transport vesicle membranes that carry hemoglobin into the DV (9, 16, 18–20). The location and structure of the lipid subphase in the parasite DV is a subject of debate. Electron microscopy observations have indicated that the lipids self-assemble into nanospheres suspended in the DV (9, 16); however, recent studies indicate the absence of suspended lipid structures and suggest that the lipids line and thicken the DV membrane (18–20). Hemozoin crystals have been observed immersed in the lipid nanospheres (9, 16), or with their basal surfaces attached to the DV membrane (18–20) and other crystal faces apparently exposed to the aqueous subphase. Whereas it has been hypothesized that hemozoin crystals nucleate on the DV membrane (17, 20), the debate on the medium, aqueous or organic, from which soluble hematin reaches the crystals and associates to them has yet to be reconciled (18–20, 31–33). To address this issue, we tested the feasibility of hematin crystallization from solutions that mimic either the aqueous phase or the lipid structures in the DV. First, we attempted to grow hematin crystals larger than the hemozoin crystals extracted from the DV of P. falciparum, which are less than 1 μm in their longest dimension (16). Our group and others have produced β-hematin crystals in aqueous solution, achieving a marginal increase in crystal size (≤3 μm), but only through the use of nonphysiological conditions (i.e., ionic strength 0.5–5 M) (34). Our attempts to grow β-hematin in aqueous solution using citric and acetate buffers (both at pH 4.8) as a surrogate for the DV produced crystals with a morphology distinctly different from that of hemozoin (34). We then used an analog to the lipid subphase in the DV, a solution of n-octanol saturated with citric buffer at pH 4.8 (details of the preparation are provided in SI Text) and referred to as citric buffer-saturated octanol, CBSO (35). In this solvent, we grew 30-μm β-hematin crystals that possessed the characteristic morphology (Fig. 1B) and powder X-ray diffraction pattern (Fig. S1C) of hemozoin crystals extracted from the DV of P. falciparum (17). Our results revealed that both the organic and aqueous components are critical for crystal growth. For instance, β-hematin crystals failed to grow in anhydrous n-octanol, which seems to suggest that H+ ions are a necessary component of the growth medium, presumably to facilitate the formation of hydrogen bonds in the crystal structure (35). Our analysis of a representative blend of lipids in the DV suggests that there is ∼8.5% (by mass) dissolved water (Fig. S2). As a second test, we used in situ atomic force microscopy (AFM) to monitor the evolution of unfinished layers on large β-hematin crystals in the presence of multiple aqueous solvents (Table S1). The unfinished layers did not grow despite the abundant growth sites presented on the curved steps (Fig. S3). A similar outcome was observed for anhydrous n-octanol (Fig. S4). However, this behavior is in direct contrast with the continuous growth of layers that were observed in CBSO solutions, as discussed below. As a third test, we determined the solubility of hematin in CBSO. Spectroscopic analyses (34) revealed that this solubility is ∼105× higher than in aqueous buffer at pH 4.8 (Fig. S6A), which is not surprising given that hematin is a hydrophobic molecule. Because crystal growth rates roughly scale with the solubility (23), this disparity in the magnitude of hematin solubility indicates that crystallization from an organic phase is a significantly faster method of heme detoxification than from an aqueous phase. Below, we use the data on hematin solubility for quantitative analyses of hematin crystallization in CBSO. An argument that is presented in the literature in favor of aqueous crystallization of hematin is the putative low solubility

Fig. 2. Generation of crystal layers. (A–D) Time-resolved in situ AFM images showing growing and dissolving islands on a (100) face at cH = 0.25 mM. Arrows indicate newly nucleated islands (I–V), islands that grow with time (I–III), an island that dissolves (IV), and an island that retains its size for the duration of observation (V). (Scale bar, 125 nm.) (E) Dependence of the critical radius of 2D nuclei Rcrit on ln(cH/ce); the solid line is the predicted trend based on the Gibbs–Thomson relation with surface free energy γ = 23 ± 5 mJ m−2 estimated using the Turnbull empirical rule (42), γ = 0.3 ΔHocryst /Ω2/3, where ΔHocryst = -37 ± 8 kJ mol−1 is the crystallization enthalpy, determined from the temperature dependence of hematin solubility in CBSO (Fig. S5); dashed lines delineate deviations due to the error in ΔHocryst and γ. (F) Rate of 2D nucleation of new layers J2D (i.e., the number of islands per unit time and area that nucleate and grow above Rcrit). The solid line is interpolated to guide the eye.

We observed a classical layer-by-layer mechanism wherein new crystal layers nucleate and grow by the attachment of solute molecules to advancing steps. Analyses of successive snapshots from AFM movies reveal that new layers may either grow (I–III in Fig. 2 A–D), dissolve (IV in Fig. 2 A–D), or retain a steady size during continuous imaging (V in Fig. 2 C and D) depending on their radius R (Movie S1). We observe a reduction in the critical radius Rcrit for island growth or dissolution with increasing hematin concentration (Fig. 2E), which is consistent with classical nucleation theory (CNT) applied to 2D crystal islands on a substrate (43). According to CNT, islands form as a result of fluctuations of the concentration of molecules on the surface. The dependence Rcrit(cH) is governed by the Gibbs–Thomson relation, according to which Rcrit = Ωγ/kBTln(cH/ce) (23) [where Ω = 0.708 nm3 is the volume of one molecule in the crystal (7); γ is the surface free energy of the layer edge; kB is the Boltzmann constant; T is temperature; cH is hematin concentration; and ce is hematin solubility in CBSO]. The correspondence between the experimentally determined Rcrit and the a priori CNT prediction in Fig. 2F indicates that the generation of new layers on growing β-hematin surfaces is governed by the thermodynamics of hematin crystallization. Analysis of in situ AFM images permits the determination of layer nucleation J2D as the number of islands that exceed Rcrit per unit area per time. According to CNT, J2D ∝ exp(–ΔGp2D =kB T), where the free-energy barrier for layer nucleation, ΔGp2D = πγRcrith, decreases with increasing cH, leading to an exponential increase of J2D with ln(cH /ce) (43, 44). Data in Fig. 2F are qualitatively consistent with this prediction, although the increase in J2D with ln(cH/ce) is weaker than this trend. This is expected because J2D is regulated by surface supersaturations that are lowered from the bulk value during growth at high deviations from equilibrium, whereas Rcrit responds to surface supersaturations equilibrated with the bulk, as evidenced by the fluctuations of surface islands around their critical size in Fig. 2 A–D. 4948 | www.pnas.org/cgi/doi/10.1073/pnas.1501023112

Upon nucleation, layers advance across the surface, merge with adjacent islands, and eventually cover the entire face (Fig. 3 A and B). The island morphology undergoes a temporal shift from an isometric to an anisotropic shape that elongates along ⇀ the c direction. The velocity v of advancing steps was determined from the average displacement Δx of steps over time by the comparison of successive AFM images, similar to those in Fig. 3 ⇀ A and B. We observed a faster step velocity along the c direction, consistent with high c/b aspect ratios of islands and bulk crystal habit, which may be attributed to the differences in kink structure (45) or density (46) along each step edge. Herein, we report ⇀ step velocity in the dominant c direction. The step velocity v exhibits a linear dependence on hematin concentration cH (Fig. 3C) if the steps are separated by more than 150 nm. This linearity indicates that hematin molecule addition to β-hematin crystals is a first-order reversible process where v(cH) reaches zero at ce and becomes negative at cH < ce, denoting step retreat due to crystal dissolution. The estimate of crystal solubility ce = 0.16 mM from in situ AFM (Fig. 3C) is in excellent agreement with bulk crystallization data (Fig. S5). The coefficient of proportionality between v and cH is referred to as the step kinetic coefficient β defined from v = βΩ(cH – ce). In turn, β is proportional to the effective first-order rate constant k for association of molecules to the steps (β = a k, where a is the molecule size). From the data in⇀ Fig. 3C, β = 4.3 μm s-1and k ≅ 104 s−1 for steps moving in the c direction. The observations in Figs. 2 and 3 indicate a strictly classical mechanism of hematin crystallization, observed for numerous other solution-grown crystals (23, 45, 47, 48). Note that in our AFM studies of growing crystal faces, we never detected the association of preassembled species, such as hematin oligomers, which would be indicative of nonclassical growth (24–26, 49).

Fig. 3. Layer growth on β-hematin {100} surfaces. (A and B) Time-resolved in situ AFM images of layer growth in solution at cH = 0.28 mM. (Scale bar, ⇀ 250 nm.) (A) Stack of six layers slightly elongated along the c direction. (B) As layers advance and merge with other layers, the anisotropy of their shape ⇀ ⇀ increases due to a faster step velocity v in the c direction relative to the b direction, i.e., v{001}/v{010} = 2.5. (C) Step velocity in the [001] direction as a function of hematin concentration cH. The solid line is a linear regression (R2 = 0.947).

Olafson et al.

The Mechanism of Action of CQ. The prevailing hypotheses regarding the suppression of hemozoin formation collectively assume that antimalarials increase cH in parasite DVs. This process can occur if drug molecules form noncrystallizable complexes with hematin in solution (i.e., hematin complexation) (51) and/or if the molecules bind to hemozoin crystals and impede the addition of solute to growing steps (i.e., crystal growth inhibition) (17). The latter mechanism was proposed by Sullivan for quinoline drugs (52) and by Leiserowitz and coworkers for artemisinin (53). To date, definitive evidence for antimalarial mode(s) of action remains elusive. Here we identify the mechanism of β-hematin crystal growth inhibition by CQ. In situ AFM measurements reveal that the addition of CQ to hematin growth solutions leads to slower step growth, fewer 2D nuclei, and more rugged step edges (Fig. 5 A–E and Movie S2). The impact of CQ at concentrations cCQ = 0–2 μM on layer generation and step propagation is summarized in Fig. 5 F–H. There is an exponential decay in J2D with increasing cCQ that is accompanied by a monotonic decrease in v (Fig. 5 G and H). We observe a complete suppression of layer nucleation and step growth at cCQ = 2 μM. These observations are consistent with the hypothesis that CQ Olafson et al.

CHEMISTRY

Concurrent with this observation, characterization of the homogeneity of supersaturated hematin solutions in CBSO by dynamic light scattering revealed the absence of any aggregates of size 1 nm and larger (35). This excludes the possibility of nonclassical crystallization of hematin. The values of J2D and v in Figs. 2 and 3, respectively, may provide insight into hematin crystallization in vivo. For instance, electron micrographs of hemozoin crystals in the parasite DV reveal that the crystals can reach thicknesses in the [100] direction of ∼100 nm within 20 h (9). Our in vitro assays suggest that this approximate rate of crystallization occurs around cH = ⇀ 0.22 mM. Indeed, at this cH the step velocity in the c direction v = 0.10 nm s−1 (Fig. 3C) and the corresponding density of steps (determined by the rate of nucleation of new layers J2D) 〈l−1〉 ≅ 0.008 nm−1. The crystal growth rate is the product r = h〈l−1〉v ≅ 8 × 10−4 nm s−1; this product is independent of the direction of step motion (50). With this r, a crystal that is 100 nm thick in the [100] direction has grown for ∼18 h, which is comparable to that of hemozoin in vivo. The closeness of the two time periods suggests that the measured layer nucleation rates and step velocities are physiologically relevant. It also suggests that the hematin concentration in vivo is close to 0.22 mM. This value is only slightly higher than the solubility of 0.16 mM and significantly lower than the potential maximum of 16 mM (i.e., the total hematin generated in an average parasite DV) (9). These two comparisons imply that hematin is incorporated into hemozoin crystals soon after its release during hemoglobin catabolism. Thus, even a moderate delay in crystallization may induce a significant accumulation of toxic hematin, leading to parasite eradication from its human host. Direct AFM observation of nucleation of new layers and their spreading by incorporation of molecules identified four classes of sites on the surface of growing hematin crystals that are important for growth and constitute potential binding sites for antimalarial drugs. These sites are illustrated in Fig. 4A for the {100} faces and their atomic structures are depicted in Fig. 4 B–G. Class 1 consists of molecularly flat surfaces that are typically located between steps (Fig. 4B). Class 2 refers to nuclei of new layers (Fig. 4C), which may comprise several molecules, and can potentially exhibit structures that are distinct from those of larger layers. Class 3 includes the kinks located along the steps (Fig. 4 D–F). There are four types of kinks, obtuse and ⇀acute, located on steps spreading the positive and negative b and ⇀ c directions. Finally, class 4 contains groups of closely spaced steps that may host large inhibitor molecules capable of bridging multiple step edges–terraces, as illustrated in Fig. 4G.

Fig. 4. Four classes of sites on the surface of growing hematin crystals to which inhibitor molecules may bind and arrest layer generation and step growth. (A) AFM image of a hematin crystal growing in CBSO. (Scale bar, 250 nm.) Numbers indicate respective classes. (A, 1) Molecularly flat segments of the growing face. (A, 2) Newly nucleated islands. (A, 3) Kinks along steps. (A, 4) Closely spaced steps. (B–G) Illustrations of the atomic structures of the four classes of sites: (B) Flat surfaces; (C) Island nuclei; (D and E) Acute and obtuse kinks on steps growing in the positive and negative ~ b and ~ c directions, respectively. (F) Profile view of an acute kink. (G) Profile view of a pair of closely spaced steps. The green spheres indicate modifiers that are physisorbed on crystal surface sites.

impedes crystallization by binding to hematin crystal surfaces, similar to inhibition mechanisms established for many biogenic and synthetic crystals in the literature (23, 54). Furthermore, the data suggest that a complexation mechanism cannot fully account for the reduced growth rates measured by in situ AFM. If, for example, we assume that CQ forms a 1:1 CQ–hematin complex, the free hematin concentration would decrease by 2 μM. Based on the v(cH) dependence in Fig. 3D, this decrease would engender a negligible reduction in v. This interpretation, however, does not rule out a potential role of CQ–hematin complexes (51) in hematin growth suppression because AFM imaging cannot identify the adsorbed inhibitor species. CQ could adsorb to any of the sites listed in Fig. 4. AFM studies reveal a site specificity of CQ for {100} terraces. The concomitant suppression in J2D and v at cCQ = 2 μM and the appearance of protrusions along advancing steps (Fig. 5 D and E) are consistent with a step-pinning (stopper) mechanism (55), i.e., CQ molecules preferentially adsorb on the crystal surface and block step propagation. This mechanism assumes that inhibitor surface coverage is governed by the dynamics of adsorption. If the separation between a pair of adsorbed inhibitors PNAS | April 21, 2015 | vol. 112 | no. 16 | 4949

Fig. 5. CQ inhibition of β-hematin growth. (A–E) Time-resolved in situ AFM images of a (100) face growing in solution at cH = 0.28 mM. (Scale bar, 250 nm.) (A and B) In the absence of CQ, the surface consists of extensive flat terraces that are progressively populated with new layers. (C–E) The introduction of 1 μM CQ at 20 min results in suppressed nucleation of new layers and corrugated step edges; both observations are consistent with a steppinning mechanism. (F ) A decrease in the rate of 2D nucleation of new layers J2D relative to that in the absence of CQ, J2D,o, with increasing CQ concentration. (Inset) Structure of CQ. (G) Effect of CQ on the step displacement Δx with increased imaging time in the presence of CQ, which is normalized by the product of interstep distance l and relative supersaturation, σ = (cH – ce)/ce. (H) A decrease in step velocity v relative to that in the absence of CQ, vo, with increasing CQ concentration.

is less than 2Rcrit, the adsorbates enforce a curvature at which the advancing step is undersaturated and growth is arrested (55). At intermediate inhibitor surface concentrations, step pinning produces rugged steps (Fig. 5E, arrow) and impedes 2D layer nucleation. The significance of this mechanism is reflected in the sensitivity of crystallization to cCQ (Fig. 5 F–H). When cCQ drops to 0.25 μM, both the generation and the growth of new layers proceed with considerable rate. This high sensitivity may be a critical factor underlying the increased resistance of P. falciparum to CQ (2). Specifically, resistant strains may have developed means to lower cCQ in the DV to levels that permit effective heme detoxification.

layers are generated by 2D nucleation and grow by the association of molecules from the solution. These mechanistic details identified four distinct classes of crystal surface sites that play crucial roles in growth and could be potentially blocked by crystallization inhibitors to prevent heme detoxification. We provide direct evidence that CQ adsorption on hematin crystal surfaces arrests heme detoxification by suppressing surface growth at concentrations as low as 2 μM. The mechanism of CQ drug action was conclusively identified: CQ adsorbs on {100} terraces between hematin growth steps and blocks step propagation. Collectively, these findings may engender a paradigm shift in the rational design of antimalarial drugs wherein the identification of molecules with site specificity for binding to hematin crystal surfaces provides both a vital criterion and platform for experimental and computational drug screening.

Conclusions Here we present, to our knowledge, the first determination of the molecular mechanism of hematin crystallization. Our results provide definitive evidence resolving several long-standing open questions on hematin crystallization and heme detoxification. The hematin solubility in organic and aqueous solvents and the morphology and dynamics of hematin crystal surfaces held in supersaturated solutions suggest that water-saturated amphiphilic organic solvents are a preferred growth environment. Time-resolved in situ AFM observations demonstrate that hematin crystallization follows a strictly classical mechanism of crystallization wherein new crystal

ACKNOWLEDGMENTS. We thank David Sullivan, Paul Roepe, and Leslie Leiserowitz for insightful discussions on hematin crystallization and hematin–drug interaction. This work was supported by the National Institutes of Health through the Nanobiology Interdisciplinary Graduate Training Program of the Gulf Coast Consortia for Quantitative Biomedical Sciences (Grant T32EB009379), National Science Foundation (Grant MCB-1244568), NASA (NNX14AD68G and NNX14AE79G), and The Welch Foundation (Grant E-1765).

1. Greenwood BM, et al. (2008) Malaria: Progress, perils, and prospects for eradication. J Clin Invest 118(4):1266–1276. 2. Eastman RT, Fidock DA (2009) Artemisinin-based combination therapies: A vital tool in efforts to eliminate malaria. Nat Rev Microbiol 7(12):864–874. 3. Sanchez CP, Dave A, Stein WD, Lanzer M (2010) Transporters as mediators of drug resistance in Plasmodium falciparum. Int J Parasitol 40(10):1109–1118. 4. Phyo AP, et al. (2012) Emergence of artemisinin-resistant malaria on the western border of Thailand: A longitudinal study. Lancet 379(9830):1960–1966. 5. Goldberg DE, Slater AF, Cerami A, Henderson GB (1990) Hemoglobin degradation in the malaria parasite Plasmodium falciparum: An ordered process in a unique organelle. Proc Natl Acad Sci USA 87(8):2931–2935. 6. Ridley RG (2002) Medical need, scientific opportunity and the drive for antimalarial drugs. Nature 415(6872):686–693.

7. Pagola S, Stephens PW, Bohle DS, Kosar AD, Madsen SK (2000) The structure of malaria pigment β-haematin. Nature 404(6775):307–310. 8. Egan TJ, Ross DC, Adams PA (1994) Quinoline anti-malarial drugs inhibit spontaneous formation of beta-haematin (malaria pigment). FEBS Lett 352(1): 54–57. 9. Egan TJ (2008) Haemozoin formation. Mol Biochem Parasitol 157(2):127–136. 10. de Villiers KA, Egan TJ (2009) Recent advances in the discovery of haem-targeting drugs for malaria and schistosomiasis. Molecules 14(8):2868–2887. 11. Corrêa Soares JB, et al. (2009) Interference with hemozoin formation represents an important mechanism of schistosomicidal action of antimalarial quinoline methanols. PLoS Negl Trop Dis 3(7):e477. 12. Sullivan DJ (2002) Theories on malarial pigment formation and quinoline action. Int J Parasitol 32(13):1645–1653.

4950 | www.pnas.org/cgi/doi/10.1073/pnas.1501023112

Olafson et al.

Olafson et al.

34. Ketchum MA, Olafson KN, Petrova EV, Rimer JD, Vekilov PG (2013) Hematin crystallization from aqueous and organic solvents. J Chem Phys 139(12):121911. 35. Olafson KN, Rimer JD, Vekilov PG (2014) Growth of large hematin crystals in biomimetic solutions. Cryst Growth Des 14(5):2123–2127. 36. Asher C, de Villiers KA, Egan TJ (2009) Speciation of ferriprotoporphyrin IX in aqueous and mixed aqueous solution is controlled by solvent identity, pH, and salt concentration. Inorg Chem 48(16):7994–8003. 37. de Villiers KA, Kaschula CH, Egan TJ, Marques HM (2007) Speciation and structure of ferriprotoporphyrin IX in aqueous solution: Spectroscopic and diffusion measurements demonstrate dimerization, but not mu-oxo dimer formation. J Biol Inorg Chem 12(1):101–117. 38. Teng HH, Dove PM, Orme CA, De Yoreo JJ (1998) Thermodynamics of calcite growth: Baseline for understanding biomineral formation. Science 282(5389):724–727. 39. Yau S-T, Vekilov PG (2000) Quasi-planar nucleus structure in apoferritin crystallization. Nature 406(6795):494–497. 40. Georgiou DK, Vekilov PG (2006) A fast response mechanism for insulin storage in crystals may involve kink generation by association of 2D clusters. Proc Natl Acad Sci USA 103(6):1681–1686. 41. Lupulescu AI, Rimer JD (2014) In situ imaging of silicalite-1 surface growth reveals the mechanism of crystallization. Science 344(6185):729–732. 42. Turnbull D (1950) Formation of crystal nuclei in liquid metals. J Appl Phys 21: 1022–1028. 43. Michely T, Krug J (2004) Islands, Mounds and Atoms. Patterns and Processes in Crystal Growth Far from Equilibrium (Springer, Heidelberg). 44. Markov IV (2003) Crystal Growth for Beginners. Foundations of Nucleation, Crystal Growth and Epitaxy (World Scientific, Singapore), 2nd Ed. 45. Vekilov PG (2007) What determines the rate of growth of crystals from solution? Cryst Growth Des 7(12):2796–2810. 46. Lovette MA, Doherty MF (2012) Multisite models to determine the distribution of kink sites adjacent to low-energy edges. Phys Rev E Stat Nonlin Soft Matter Phys 85(2 Pt 1):021604. 47. De Yoreo JJ, Vekilov PG (2003) Principles of crystal nucleation and growth. Biomineralization. Rev Mineral Geochem 54:57–93. 48. Chernov AA (1989) Growth of crystals from solutions. Contemp Phys 30:251–276. 49. Song R-Q, Cölfen H (2010) Mesocrystals—ordered nanoparticle superstructures. Adv Mater 22(12):1301–1330. 50. Vekilov PG, Kuznetsov YG, Chernov AA (1992) Interstep interaction in solution growth; (101) ADP face. J Cryst Growth 121:643–655. 51. Gorka AP, de Dios A, Roepe PD (2013) Quinoline drug-heme interactions and implications for antimalarial cytostatic versus cytocidal activities. J Med Chem 56(13): 5231–5246. 52. Sullivan DJ, Jr, Gluzman IY, Russell DG, Goldberg DE (1996) On the molecular mechanism of chloroquine’s antimalarial action. Proc Natl Acad Sci USA 93(21): 11865–11870. 53. Solomonov I, et al. (2007) Crystal nucleation, growth, and morphology of the synthetic malaria pigment β-hematin and the effect thereon by quinoline additives: The malaria pigment as a target of various antimalarial drugs. J Am Chem Soc 129(9): 2615–2627. 54. Elhadj S, De Yoreo JJ, Hoyer JR, Dove PM (2006) Role of molecular charge and hydrophilicity in regulating the kinetics of crystal growth. Proc Natl Acad Sci USA 103(51):19237–19242. 55. Cabrera N, Vermileya DA (1958) The growth of crystals form solution. Growth and Perfection of Crystals, eds Doremus RH, Roberts BW, Turnbull D (Wiley, New York).

PNAS | April 21, 2015 | vol. 112 | no. 16 | 4951

CHEMISTRY

13. Haynes RK, et al. (2011) A partial convergence in action of methylene blue and artemisinins: Antagonism with chloroquine, a reversal with verapamil, and an insight into the antimalarial activity of chloroquine. ChemMedChem 6(9):1603–1615. 14. Kappe SHI, Vaughan AM, Boddey JA, Cowman AF (2010) That was then but this is now: Malaria research in the time of an eradication agenda. Science 328(5980):862–866. 15. Bohle DS, Dinnebier RE, Madsen SK, Stephens PW (1997) Characterization of the products of the heme detoxification pathway in malarial late trophozoites by X-ray diffraction. J Biol Chem 272(2):713–716. 16. Pisciotta JM, et al. (2007) The role of neutral lipid nanospheres in Plasmodium falciparum haem crystallization. Biochem J 402(1):197–204. 17. Weissbuch I, Leiserowitz L (2008) Interplay between malaria, crystalline hemozoin formation, and antimalarial drug action and design. Chem Rev 108(11):4899–4914. 18. Kapishnikov S, et al. (2012) Oriented nucleation of hemozoin at the digestive vacuole membrane in Plasmodium falciparum. Proc Natl Acad Sci USA 109(28):11188–11193. 19. Kapishnikov S, et al. (2012) Aligned hemozoin crystals in curved clusters in malarial red blood cells revealed by nanoprobe X-ray Fe fluorescence and diffraction. Proc Natl Acad Sci USA 109(28):11184–11187. 20. Kapishnikov S, et al. (2013) Digestive vacuole membrane in Plasmodium falciparuminfected erythrocytes: Relevance to templated nucleation of hemozoin. Langmuir 29(47):14595–14602. 21. Stranski IN (1928) Zur theorie des Kristallwachstums. Z phys Chem 136:259–278. 22. Burton WK, Cabrera N, Frank FC (1951) The growth of crystals and equilibrium structure of their surfaces. Phil Trans R Soc London Ser A 243:299–360. 23. Chernov AA (1984) Modern Crystallography III, Crystal Growth (Springer, Berlin). 24. Banfield JF, Welch SA, Zhang H, Ebert TT, Penn RL (2000) Aggregation-based crystal growth and microstructure development in natural iron oxyhydroxide biomineralization products. Science 289(5480):751–754. 25. Li D, et al. (2012) Direction-specific interactions control crystal growth by oriented attachment. Science 336(6084):1014–1018. 26. Van Driessche AES, et al. (2012) The role and implications of bassanite as a stable precursor phase to gypsum precipitation. Science 336(6077):69–72. 27. Dinio T, Gorka AP, McGinniss A, Roepe PD, Morgan JB (2012) Investigating the activity of quinine analogues versus chloroquine resistant Plasmodium falciparum. Bioorg Med Chem 20(10):3292–3297. 28. Gorka AP, et al. (2013) Cytostatic versus cytocidal activities of chloroquine analogues and inhibition of hemozoin crystal growth. Antimicrob Agents Chemother 57(1): 356–364. 29. Weissbuch I, Lahav M, Leiserowitz L (2003) Toward stereochemical control, monitoring and understanding of crystal nucleation. Cryst Growth Des 3:125–150. 30. Gligorijevic B, Bennett T, McAllister R, Urbach JS, Roepe PD (2006) Spinning disk confocal microscopy of live, intraerythrocytic malarial parasites. 2. Altered vacuolar volume regulation in drug resistant malaria. Biochemistry 45(41):12411–12423. 31. Hoang AN, Ncokazi KK, de Villiers KA, Wright DW, Egan TJ (2010) Crystallization of synthetic haemozoin (beta-haematin) nucleated at the surface of lipid particles. Dalton Trans 39(5):1235–1244. 32. Hoang AN, Sandlin RD, Omar A, Egan TJ, Wright DW (2010) The neutral lipid composition present in the digestive vacuole of Plasmodium falciparum concentrates heme and mediates β-hematin formation with an unusually low activation energy. Biochemistry 49(47):10107–10116. 33. Ambele MA, Egan TJ (2012) Neutral lipids associated with haemozoin mediate efficient and rapid β-haematin formation at physiological pH, temperature and ionic composition. Malar J 11:337.

Supporting Information Olafson et al. 10.1073/pnas.1501023112 SI Text Materials and Methods Preparation of Growth Solutions for Hematin Crystallization. Materials. The following reagents were purchased from Sigma-

Aldrich: porcine hematin, n-octanol (anhydrous, ≥99%), citric acid (anhydrous), sodium hydroxide (anhydrous reagent-grade pellets, ≥98%), and CQ diphosphate salt (≥98%). All materials were used as received unless otherwise noted. Deionized (DI) water was produced by a Millipore reverse osmosis–ion exchange system (RiOs-8 Proguard 2–MilliQ Q-guard). Citric buffer and CBSO. Before solution preparation, all glassware was thoroughly cleaned with 0.1 M NaOH to remove residual hematin. The glass containers were then washed with soap, rinsed with DI water, and dried in air for about 1 d. To prepare citric buffer at pH 4.80, citric acid was dissolved at 50 mM in DI water and titrated by 0.10 M NaOH to desired pH with continuous stirring. The buffer was stored at a maximum shelf life of 1 mo. Before each experiment, the buffer pH was verified. To prepare CBSO, 5 mL of citric buffer stock solution at pH 4.80 was placed in a precleaned 40-mL glass vial and 10–20 mL of n-octanol was added. The resulting two-phase solution was sealed with a polytetrafluoroethylene (PTFE) storage cap and allowed to equilibrate without stirring at 23 °C for 30 min. The denser citric buffer occupied the bottom layer and the n-octanol formed an orthogonal phase on top. The latter (top layer) was carefully removed with a pipette far from the interface to avoid resuspension of the phases. Fresh CBSO solutions were prepared before every crystal synthesis and in situ AFM experiment. The extinction coefficients of hematin and CQ in CBSO. Solid amorphous hematin does not dissolve fully in CBSO; therefore, a layered solution was utilized to determine the extinction coefficient. A known mass of amorphous hematin was added to a basic aqueous solution of 0.1 M NaOH, where it readily dissolves. This stock solution was diluted to 20 different concentrations and the absorbance spectrum of each sample was measured in the visible wavelength range. We obtained e = 3.86 ± 0.06 cm−1 mM−1 for hematin in 0.1 M NaOH using the Beer Lambert Law at λ = 607 nm. Various concentrations of hematin in 0.1 M NaOH solution were placed into contact with a two-phase CBSO solution where the amount of hematin partitioned into the organic phase was identified spectrophotometrically. To a solution of CBSO, we fully dissolved (as verified by microscopic observation) known masses of CQ diphosphate. The dissolution of CQ in CBSO is slow; therefore, 5 d at constant stirring was required to dissolve this compound. The stock solutions were diluted to 10 different concentrations. For each concentration, the absorbance spectrum in the wavelength range of 200–800 nm was recorded. A linear trend in absorbance as a function of the respective concentrations of hematin cH and CQ cCQ was observed at the wavelengths of the absorbance peaks. For hematin, we chose the peak at 594 nm, for which the absorbance(cH) data fitted with the Beer–Lambert law and yielded e = 3.1 ± 0.3 cm−1·mM−1. For CQ, we chose the highest peak at 331 nm, for which the absorbance (cCQ) data yielded e = 7.5 ± 0.2 cm−1·mM−1. Supersaturated hematin solutions in CBSO. To prepare supersaturated hematin solutions in CBSO for crystal growth experiments, we placed 8 mL CBSO and an amount of hematin powder sufficient for a 6 mM solution in a 20-mL glass vial. As previously shown in Olafson et al. (1), the hematin in this mixture is amorphous. The unsealed vial was kept at 39.5 ± 0.2 °C in the absence of direct light. Results discussed below indicate that the hematin solubility Olafson et al. www.pnas.org/cgi/content/short/1501023112

in CBSO at this temperature is 0.32 mM. Solution samples were periodically extracted from these solutions by pipette. Each aliquot was filtered through a 0.22-μm polyvinylidene fluoride (PVDF) membrane, and the hematin concentration was measured spectrophotometrically using the previously determined extinction coefficient. The concentration of the working solutions was adjusted to 0.1–0.3 mM by the addition of CBSO. The solutions were used in AFM experiments within 3 h of preparation. CBSO solutions containing hematin may exhibit two heterogeneities: hematin aggregates and water nanodroplets (1). To test for heterogeneity, we used dynamic light scattering (DLS) with an ALV-5000/EPP device equipped with a laser (λ = 628 nm). Detailed protocols for DLS measurements and data analysis are discussed in Ketchum et al. (2). The recorded correlation functions indicated the absence of any objects (molecules, aggregates, or droplets) of size larger than 1 nm. From this study, we concluded that the solution is homogeneous, i.e., hematin aggregation and water phase separation are suppressed. Dissolution of CQ. To dissolve CQ in biomimetic solutions we placed solid CQ phosphate power and 10 mL CBSO in a clean 20-mL glass vial. The vial was sealed with a PTFE cap and incubated at 22.6 ± 0.2 °C in the absence of light. Aliquots of this solution were filtered through a 0.22-μm PVDF membrane and the CQ concentration was determined spectrophotometrically using the previously determined extinction coefficient. The CQ solutions were mixed with hematin-containing solutions to desired concentrations of the two solutes and were used in AFM experiments within 10 min of preparation. Preparation of Crystals for AFM Observation. Crystallization. β-Hematin crystals of size 10–30 μm were prepared

by placing ∼2 mM hematin powder in direct contact with 5 mL freshly prepared CBSO held in a 20-mL glass vial (Fig. S1A) (1). The vial was sealed with a PTFE storage cap and incubated at 66.4 ± 0.5 °C for 2 d followed by 1-d incubation at room temperature (22.6 ± 0.2 °C). At the end of incubation, crystals of size 10–50 μm were observed at the bottom of the vial (Fig. S1B). Mounting of crystals for AFM observation. For AFM imaging, hematin crystals were grown on 12-mm-round glass coverslips (Ted Pella). Before crystal growth, the coverslip was scratched near the center using a diamond cutter and placed in the vial containing hematin powder and CBSO with the scratched side exposed to the solution. The scratched surface presented an interface for heterogeneous nucleation and growth. Indeed, we observed that several crystals grew attached to the coverslip, mostly at the scratched grooves. We removed the coverslip, rinsed it with DI water and ethanol, and dried it under a gentle flow of air. All prepared crystals were used within 1–2 d of drying. Validation of β-hematin crystal structure by powder X-ray diffraction. The crystals grown for AFM studies were removed from the glass coverslips and washed with ethanol to remove the supernatant. The solid was dried at room temperature (22.6 ± 0.2 °C) for 48 h. Powder X-ray diffraction (XRD) patterns were collected on a Siemens D5000 X-ray diffractometer with CuKα radiation (40 kV, 40 mA, λ = 1.54 Å). The experimental pattern was compared with a model pattern of β-hematin that was simulated using the software package Diamond. The d spacings of as-synthesized β-hematin seed crystals coincide with those from the model β-hematin, which is based on structure coordinates from the Cambridge Structural Database. The XRD patterns are provided in Fig. S1C. For comparison we also include an amorphous pattern (porcine hematin as received from the manufacturer), 1 of 6

which was reported in our previous work (1). The powder XRD pattern of the latter reveals amorphous material, as indicated by the lack of distinct Bragg peaks. Atomic Force Microscopy. Imaging modes and scanning parameters. We used a Multimode

atomic force microscope (Nanoscope IV) from Digital Instruments. AFM images were captured in either contact or tapping mode using Olympus TR800PSA probes (Silicon nitride, Cr–Au coated, with spring constant 0.15 N/m). We collected images in height and amplitude (or deflection) modes with scan rates between 2 and 2.5 s−1 and 256 scan lines per image. Contact mode images were used in select cases to help visualize surface features; however, all in situ measurements were performed in tapping mode to minimize the possibility of disturbing solute and/or growth modifier attachment to crystal surfaces. Tapping-mode images were collected at a tapping frequency of 32 kHz. To judge whether continuous AFM scanning affected the dynamics of surface growth, we periodically performed a standard test. We continuously imaged a surface area using typical scanning parameters for about 1 h, after which we increased the scan size so that the previously viewed area is fully incorporated within the new image. The continuity of surface features between the previously imaged area and its periphery was interpreted as evidence for the lack of interference between the AFM tip and the nucleation and growth of layers on the crystal surface. Moreover, multiple areas of different crystals were assessed at periodic times to confirm that the features selected for in situ AFM analysis were representative of the entire sample. The temperature in the AFM fluid cell. The temperature of the growth solution within the AFM fluid cell during in situ crystal growth monitoring was measured with a copper-constantan thermocouple connected to a temperature controller (SE5010, Marlow Industries Inc.). The thermocouple was calibrated using a crushed ice–DI water bath. The bath was allowed to equilibrate for 30 min, after which the water freezing point was verified at several locations with an accurate (±0.1 °C) mercury thermometer. The thermocouple tip was secured to the thermometer. The set point of the controller was adjusted to read 0.0 ± 0.1 °C. The thermocouple was embedded in a brass disc positioned right under the AFM sample. The liquid cell was sealed with a silicon O-ring and loaded with CBSO. To replicate the experimental conditions, we continuously imaged for 3 h a 2 × 2-μm2 area at a scanning rate of 2.52 s−1. The temperature in the fluid cell reached a steady value of 27.8 ± 0.1 °C within 15 min of imaging. In situ AFM imaging of β-hematin surface growth in real time. We mounted the coverslips with β-hematin crystals on 15-mm metal AFM specimen discs. We placed white paper circles between the discs and the coverslips to increase the visibility of the crystals in the optical microscope during AFM cantilever alignment. The viscosity of CBSO (7.6 mPa s) increased the lateral drift during imaging, which significantly reduced the image quality relative to higher-resolution imaging environments, i.e., tapping-mode images taken in aqueous solvents and contact mode images taken in CBSO. To address this issue, we used vacuum grease (Dow Chemicals) to lubricate the silicone O-ring that seals the AFM fluid cell and the periphery of the glass coverslip in contact with it. To test if the grease affects the monitored processes of surface nucleation and growth, we conducted experiments with and without lubrication and found that lubrication did not affect the surface morphology or the rate of generation and spreading of layers. We loaded the AFM fluid cell with working solution of hematin in CBSO with concentration 0.1–0.3 mM. For this, we used disposable polypropylene 1-mL syringes, which are tolerant of the organic solvent. After loading, this solution was equilibrated for 15–30 min. The working solution was renewed every ∼30 min to maintain the desired hematin concentration. Olafson et al. www.pnas.org/cgi/content/short/1501023112

AFM Data Analysis. AFM image processing. We processed height and amplitude images

by first- or third-order flattening, or by plane fitting. Select images with sizes below 1 μm were processed with a 2D fast Fourier transform filter to remove periodic noise. Three-dimensional height plots were adjusted with a Gaussian filter using a filter size of 2.56 nm. No low-pass or median filters were applied to the AFM images. Determination of the radius of the 2D nucleus of new layers. The critical radius Rcrit for layer nucleation is defined as the threshold size above which an island has a higher probability to grow. Islands of size R < Rcrit are more likely to dissolve. In time-resolved sequences of in situ AFM images captured in height mode, we monitored the size evolution of all newly generated islands and classified them as growing or dissolving. The largest sizes reached by dissolving islands and the threshold above which all islands grew were averaged to yield Rcrit. Determination of the nucleation rate of new layers. The rate of layer nucleation J2D was determined from time-resolved sequences of in situ AFM images captured in height mode. We counted the number of new islands that appeared on the monitored surface for the time between two images. We monitored the evolution of each island in subsequent images and eliminated those that dissolve. Thus, only islands that grew beyond Rcrit were considered. This number was scaled with the imaged area to yield J2D (nm−2·s−1). Determination of step velocity and interstep distances. The step velocity v was determined from time-resolved sequences of in situ AFM images captured in height mode. Typical times between images were ∼1–2 min. First, the crystal edge defined by the intersection of the studied ð100Þ surface and either of the adjacent two f010g faces was located on the image (at a larger scanning area). From ⇀ ⇀ these images we identified the crystallographic b and c directions, which are at 90.2° and 0° relative to the crystal edge, respectively. To account for the effects of lateral drift, a reference point was located in the image. This was either a surface defect or feature that did not shift over the course of imaging. For statistical analysis, a minimum of 3 steps and as many as 10 steps were tracked in an image sequence. Only single steps, which exhibit a measured step height of 1.17 ± 0.07 nm, were ⇀consid⇀ ered. In each image, the step displacements along the b and c directions with respect to the reference point were measured. For each step, the displacement was plotted as a function of time ⇀ ⇀ and the slope of this dependence yielded v in either b or c directions (reported in units of nm s−1). Two averages, v〈010〉 and ⇀ in the c v〈001〉, were calculated. Because the step velocity ⇀ directions was ∼2.5× faster than that in the b directions, only v〈001〉 is reported. This process was repeated for each hematin and CQ concentration. The ⇀interstep distances ℓ were measured ⇀ for each image along the b and c directions at 90.2° and 0° relative to the crystal edge, respectively. Extended Discussions Using CBSO as a Mimic of the Lipid Structures in the DV. Recent analyses indicate that the lipid structures in the DV of P. falciparum consist of monopalmitic, monostearic, dipalmitic, dioleic, and dilinoleic glycerols at molar ratios 2:4:1:1:1 (3, 4). Because this mixture is solid at room temperature (it is liquid at 37 °C), octanol has been suggested as a convenient biomimetic model (5, 6). Octanol is an amphiphilic molecule with an aliphatic tail of medium length and a polar head (-OH), which is smaller than the glycerol ester functional groups of the mono- and diglyceride lipids. Because the lipid structures are in contact with the aqueous environment of the DV, it is feasible that they are saturated with water. To test this hypothesis, we determined the water solubility in a mixture of lipids. We prepared samples of the five-lipid mixture at the reported molar ratios and placed them in contact with equal volumes of DI water at 60 °C without stirring. The lipid mixture was kept in contact with water for 1 h. 2 of 6

Aliquots of the upper lipid layer were removed by pipetting. We recorded the mass fraction of evaporated solvent from 3-mg samples using a TA Instruments SDT-Q600 for thermal gravimetric analysis (TGA) and under nitrogen flowing at 50 mL min−1. We used three different ramp rates (0.1 °C, 0.3 °C, and 1.0 °C min−1) from 25 °C to 80 °C and then a ramp rate of 0.1 °C min−1 from 80 °C to 120 °C (Fig. S2). Independent of the temperature ramp rate, the lipid samples reached a constant mass between 70 °C and 85 °C that did not increase for temperatures up to 120 °C. Because these temperatures are significantly lower than the boiling points of the lipids, we conclude that the evaporated solvent component is water. The three traces in Fig. S2 indicate that the average amount of dissolved water in the lipid mixture is 8.5 ± 0.5% mass, corresponding to 4.2 ± 0.3 mol kg−1. Water solubility in n-octanol is somewhat lower, 2.70 mol kg−1 (7). Crystallization in Alternative Solvents. Aqueous solvents. Previous AFM observations of β-hematin crystals

grown from aqueous solutions at pH 4.80 revealed that they have unusual macroscopic and mesoscopic morphological features (2), which are inconsistent with the shapes and surface features of biological hemozoin. This discrepancy cannot be attributed to impurity effects because impurities are abundant in vivo. Hence, one would expect rougher crystal shapes of biological hemozoin, which is in stark contrast with the actual observation (2). Crystals with smooth, faceted features did not grow under any of the tested conditions (2). Here we used in situ AFM in multiple aqueous solvents to monitor the behavior of the large β-hematin crystals discussed above (these crystals readily grow from hematin solutions in CBSO). To mimic the aqueous environment in the DV of P. falciparum, we prepared solutions at low ionic strength. To simulate conditions used in previous studies of hematin crystallization (8), we also probed high ionic strengths. In addition, we tested several pH values in which growth solutions were prepared by adding hematin to the crystallization solution using two methods: dissolving in 0.1 M NaOH and titrating to the desired pH, or placing solid hematin in the solvent, analogous to the CBSO solution preparation. We tested aqueous hematin solution saturated with n-octanol to bring them closer to the composition in the parasite DV. We examined both filtered and unfiltered solutions to observe whether large aggregates forming in the solution might act as nonclassical growth units and/or sites for heterogeneous nucleation. For each solution composition we imaged the crystal surface for over 48 h in tapping mode. No surface processes leading to crystal growth were observed for any of the aforementioned solutions. In acidic solutions, immobile single steps were initially observed on the basal face, but they became obscured with increased time of exposure to the solution due to the deposition of an apparent coating (Fig. S3). In basic solutions (i.e., high solubility of hematin) the steps retreated, thereby indicating dissolution of the crystals. We did not detect any dynamic events that could allude to nonclassical mechanisms (e.g., precursor aggregate deposition and restructuring to become part of the crystal). A summary of all tested growth solution compositions and the observed surface behaviors in AFM studies is provided in Table S1. Anhydrous n-octanol. AFM observations revealed that in anhydrous octanol, in which the solubility of hematin is even greater than its value measured in CBSO, steps retracted at hematin concentrations lower than the solubility, which is consistent with crystal dissolution. At concentrations higher than the solubility, continuous AFM scanning for 48 h revealed no step growth (Fig. S4). 1. Olafson KN, Rimer JD, Vekilov PG (2014) Growth of large hematin crystals in biomimetic solutions. Cryst Growth Des 14(5):2123–2127. 2. Ketchum MA, Olafson KN, Petrova EV, Rimer JD, Vekilov PG (2013) Hematin crystallization from aqueous and organic solvents. J Chem Phys 139(12):121911.

Olafson et al. www.pnas.org/cgi/content/short/1501023112

Determination of the hematin solubility in CBSO and the crystallization enthalpy. To determine the solubility of hematin in CBSO, we added

2–5 mM hematin powder to 5 mL CBSO held in 20-mL glass vials that were subsequently sealed with Parafilm. Sets of three vials were stored at four temperatures: 5 °C, 25 °C, 37 °C, and 45 °C. A 300-μL aliquot was removed weekly from each of the 12 solutions, diluted with CBSO, filtered through a 0.22-μm polyethersulfone (PES) filter, and the concentration of dissolved hematin was determined spectrophotometrically. This procedure was repeated until the concentrations in each vial reached a plateau, defined by three consecutive readings of similar value. Microscopic observations revealed that the initial amorphous hematin powder appeared to fully dissolve and β-hematin crystals with typical rodlike habit (9) had formed. The final steady-state concentrations were averaged over the three samples at each temperature. The resulting mean was used as the solubility ce of hematin in CBSO with respect to β-hematin crystals. The data for the four tested temperatures are plotted in Fig. S5, where they are compared with a determination using AFM. Solubility in AFM studies was determined as the concentration of hematin at the point of zero step velocity (i.e., the x intercept in Fig. 3C). In Fig. S6A we compare the solubility of hematin in CBSO at 25 °C to that in aqueous citric buffer at the same temperature and pH 4.8 (for details see ref. 2). In the crystallization equilibrium hematin(solution) ⇆ hematin (crystal), the product is a solid phase and has activity of one. We assume that the activity of the soluble hematin is equal to its concentration. Hence, the equilibrium constant for crystallization is Kcryst = ce-1. With this relation, the temperature dependence of solubility is given by the classical van ’t Hoff equation, 0 ΔHcryst ∂lnce = ; ∂ð1=TÞ R

[S1]

0 where ΔHcryst is the crystallization enthalpy and R is the universal gas constant. The slope of the dependence lnce(1/T) in Fig. S5B 0 yields ΔHcryst = −37 ± 8 kJ mol−1.

The CQ Solubility in CBSO and Citric Buffer at pH 4.80. The solubility of CQ in CBSO and aqueous citric buffer at pH 4.80 was measured by adding incremental amounts of the drug to 5 mL of solvent. Undissolved powder remained at the bottom of the glass vial. The solutions were then equilibrated at the following temperatures: 5 °C, 25 °C, 37 °C, and 45 °C. Each solution was prepared in triplicate for statistical analysis. Once a week, small aliquots of known volume were removed from each of the 12 solutions and diluted with the respective solvent. The diluted solutions were then filtered through a 0.22-μm PES filter and analyzed spectrophotometrically to determine the concentration of dissolved drug. This procedure was repeated until the measured concentrations reached a constant value. The measured solubilities were in the range 0.09–0.35 mM in CBSO and 0.54–71 M in the aqueous buffer, depending on the temperature. CQ solubility at 25 °C in CBSO and aqueous citric buffer are compared in Fig. S6B. All experiments were conducted using the CQ reagent as received, i.e., with minor impurity (